I tried to test China’s newest ChatGPT rival but couldn’t get in — and it exposed the next big problem facing the major AI players

China’s latest ChatGPT rival, Kimi, is now available for download on the Apple App Store, so I downloaded it on my iPhone, intending to test it against ChatGPT, Claude and Gemini. I had the prompts ready; I had the comparison planned. Then Kimi presented me with a very modern problem that I wasn’t expecting: too many people were already trying to talk to it.

This is not just app-launch inconvenience. It captures one of the biggest problems in AI right now: being clever is one thing, but the best models are only useful if companies have enough servers and GPUs to allow people to actually use them.

In our recent Kimi review we said: "Kimi delivers impressive performance for coding, document analysis, and multi-step agentic tasks, especially considering its price point." Kimi may be positioned as a serious open-weight rival to US models, but if the first ordinary-user experience is that they're stuck queuing at the door, how useful is it really?

What is Kimi?

Do you remember back in 2025, when DeepSeek-R1 was released and the internet went wild for Chinese AI models? It was a game-changing moment, because it immediately became obvious that we weren’t beholden to the Silicon Valley giants like OpenAI, Google and Anthropic for AI. China could create AI models too — and, it turned out, they were a lot cheaper to run, and almost as good.

While DeepSeek was arguably the first Chinese AI model to really break through into the mainstream, Kimi is the latest. Its story is slightly different, because it’s a giant open-weight model that’s not just cheap and technically impressive, but directly useful for coding, spreadsheets, knowledge work, and long-context tasks with agent-style workflows.

Open weight means the finished AI model is available for others to download and run, rather than only being accessible through the company’s own app or website. That is not the same as being fully open source, of course. Open source would imply a much more complete release of the code, training process, data details and license freedoms needed to understand, rebuild and modify the model from the ground up.

The data squeeze

I’d love to be able to open my phone, try a serious Chinese AI model, and get something genuinely competitive in minutes. But without the capacity to serve the millions of people who pile onto a hyped new AI model, raw intelligence only gets you so far.

That is why Kimi’s error message matters, and it points to the issue currently reshaping the entire AI industry. In the past week alone, Reuters reported that Nvidia is discussing huge financing guarantees to help OpenAI lease a proposed 10-gigawatt data center in Ohio, while other tech giants are racing to secure more data-center capacity of their own.

In other words, the next stage of the AI race may not be decided purely by benchmark scores or clever demos. It may be decided by power, servers, GPUs and the ability to keep a popular model online when millions of people suddenly want to use it.

Kimi may be impressive, but my first experience with it wasn’t about intelligence; it was about access. In 2026, the AI race is no longer just about who can build the best model. It’s about who can keep the thing online when everyone wants to use it.

Graham is the Senior Editor for AI at TechRadar. With over 25 years of experience in both online and print journalism, Graham has worked for various market-leading tech brands including Computeractive, PC Pro, iMore, MacFormat, Mac|Life, Maximum PC, and more. He specializes in reporting on everything to do with AI and has appeared on BBC TV shows like BBC One Breakfast and on Radio 4 commenting on the latest trends in tech. Graham has an honors degree in Computer Science and spends his spare time podcasting and blogging.